Rúben Almeida
commited on
Commit
·
3081464
1
Parent(s):
edebf90
Update version of requirements
Browse files- dto.py +45 -0
- main.py +4 -45
- requirements.txt +3 -3
- tests/test_awq.py +5 -3
dto.py
ADDED
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from abc import ABC
|
2 |
+
from typing import Optional
|
3 |
+
from pydantic import BaseModel, Field
|
4 |
+
|
5 |
+
|
6 |
+
### DTO Definitions
|
7 |
+
class QuantizationConfig(ABC, BaseModel):
|
8 |
+
pass
|
9 |
+
class ConvertRequest(ABC, BaseModel):
|
10 |
+
hf_model_name: str
|
11 |
+
hf_tokenizer_name: Optional[str] = Field(None, description="Hugging Face tokenizer name. Defaults to hf_model_name")
|
12 |
+
hf_token: Optional[str] = Field(None, description="Hugging Face token for private models")
|
13 |
+
hf_push_repo: Optional[str] = Field(None, description="Hugging Face repo to push the converted model. If not provided, the model will be downloaded only.")
|
14 |
+
### -------
|
15 |
+
|
16 |
+
### Quantization Configurations
|
17 |
+
class AWQQuantizationConfig(QuantizationConfig):
|
18 |
+
zero_point: Optional[bool] = Field(True, description="Use zero point quantization")
|
19 |
+
q_group_size: Optional[int] = Field(128, description="Quantization group size")
|
20 |
+
w_bit: Optional[int] = Field(4, description="Weight bit")
|
21 |
+
version: Optional[str] = Field("GEMM", description="Quantization version")
|
22 |
+
|
23 |
+
class GPTQQuantizationConfig(QuantizationConfig):
|
24 |
+
pass
|
25 |
+
|
26 |
+
class GGUFQuantizationConfig(QuantizationConfig):
|
27 |
+
pass
|
28 |
+
class AWQConvertionRequest(ConvertRequest):
|
29 |
+
quantization_config: Optional[AWQQuantizationConfig] = Field(
|
30 |
+
default_factory=lambda: AWQQuantizationConfig(),
|
31 |
+
description="AWQ quantization configuration"
|
32 |
+
)
|
33 |
+
|
34 |
+
class GPTQConvertionRequest(ConvertRequest):
|
35 |
+
quantization_config: Optional[GPTQQuantizationConfig] = Field(
|
36 |
+
default_factory=lambda: GPTQQuantizationConfig(),
|
37 |
+
description="GPTQ quantization configuration"
|
38 |
+
)
|
39 |
+
|
40 |
+
class GGUFConvertionRequest(ConvertRequest):
|
41 |
+
quantization_config: Optional[GGUFQuantizationConfig] = Field(
|
42 |
+
default_factory=lambda: GGUFQuantizationConfig(),
|
43 |
+
description="GGUF quantization configuration"
|
44 |
+
)
|
45 |
+
### -------
|
main.py
CHANGED
@@ -1,13 +1,12 @@
|
|
1 |
import zipfile
|
2 |
-
from
|
3 |
-
from typing import Optional, Union
|
4 |
from awq import AutoAWQForCausalLM
|
5 |
-
from pydantic import BaseModel, Field
|
6 |
from transformers import AutoTokenizer
|
7 |
from tempfile import NamedTemporaryFile
|
8 |
from contextlib import asynccontextmanager
|
9 |
from fastapi import FastAPI, HTTPException
|
10 |
from fastapi.responses import RedirectResponse, FileResponse
|
|
|
11 |
|
12 |
### FastAPI Initialization
|
13 |
@asynccontextmanager
|
@@ -17,46 +16,6 @@ async def lifespan(app:FastAPI):
|
|
17 |
app = FastAPI(title="Huggingface Safetensor Model Converter to AWQ", version="0.1.0", lifespan=lifespan)
|
18 |
### -------
|
19 |
|
20 |
-
### DTO Definitions
|
21 |
-
class QuantizationConfig(ABC, BaseModel):
|
22 |
-
pass
|
23 |
-
class ConvertRequest(ABC, BaseModel):
|
24 |
-
hf_model_name: str
|
25 |
-
hf_tokenizer_name: Optional[str] = Field(None, description="Hugging Face tokenizer name. Defaults to hf_model_name")
|
26 |
-
hf_token: Optional[str] = Field(None, description="Hugging Face token for private models")
|
27 |
-
hf_push_repo: Optional[str] = Field(None, description="Hugging Face repo to push the converted model. If not provided, the model will be downloaded only.")
|
28 |
-
### -------
|
29 |
-
|
30 |
-
### Quantization Configurations
|
31 |
-
class AWQQuantizationConfig(QuantizationConfig):
|
32 |
-
zero_point: Optional[bool] = Field(True, description="Use zero point quantization")
|
33 |
-
q_group_size: Optional[int] = Field(128, description="Quantization group size")
|
34 |
-
w_bit: Optional[int] = Field(4, description="Weight bit")
|
35 |
-
version: Optional[str] = Field("GEMM", description="Quantization version")
|
36 |
-
|
37 |
-
class GPTQQuantizationConfig(QuantizationConfig):
|
38 |
-
pass
|
39 |
-
|
40 |
-
class GGUFQuantizationConfig(QuantizationConfig):
|
41 |
-
pass
|
42 |
-
class AWQConvertionRequest(ConvertRequest):
|
43 |
-
quantization_config: Optional[AWQQuantizationConfig] = Field(
|
44 |
-
default_factory=lambda: AWQQuantizationConfig(),
|
45 |
-
description="AWQ quantization configuration"
|
46 |
-
)
|
47 |
-
|
48 |
-
class GPTQConvertionRequest(ConvertRequest):
|
49 |
-
quantization_config: Optional[GPTQQuantizationConfig] = Field(
|
50 |
-
default_factory=lambda: GPTQQuantizationConfig(),
|
51 |
-
description="GPTQ quantization configuration"
|
52 |
-
)
|
53 |
-
|
54 |
-
class GGUFConvertionRequest(ConvertRequest):
|
55 |
-
quantization_config: Optional[GGUFQuantizationConfig] = Field(
|
56 |
-
default_factory=lambda: GGUFQuantizationConfig(),
|
57 |
-
description="GGUF quantization configuration"
|
58 |
-
)
|
59 |
-
### -------
|
60 |
|
61 |
@app.get("/", include_in_schema=False)
|
62 |
def redirect_to_docs():
|
@@ -102,11 +61,11 @@ def convert(request: AWQConvertionRequest)->Union[FileResponse, dict]:
|
|
102 |
raise HTTPException(status_code=500, detail="Failed to convert model")
|
103 |
|
104 |
@app.post("/convert_gpt_q", response_model=None)
|
105 |
-
def convert_gpt_q(request:
|
106 |
raise HTTPException(status_code=501, detail="Not implemented yet")
|
107 |
|
108 |
@app.post("/convert_gguf", response_model=None)
|
109 |
-
def convert_gguf(request:
|
110 |
raise HTTPException(status_code=501, detail="Not implemented yet")
|
111 |
|
112 |
@app.get("/health")
|
|
|
1 |
import zipfile
|
2 |
+
from typing import Union
|
|
|
3 |
from awq import AutoAWQForCausalLM
|
|
|
4 |
from transformers import AutoTokenizer
|
5 |
from tempfile import NamedTemporaryFile
|
6 |
from contextlib import asynccontextmanager
|
7 |
from fastapi import FastAPI, HTTPException
|
8 |
from fastapi.responses import RedirectResponse, FileResponse
|
9 |
+
from .dto import AWQConvertionRequest, GGUFConvertionRequest, GPTQConvertionRequest
|
10 |
|
11 |
### FastAPI Initialization
|
12 |
@asynccontextmanager
|
|
|
16 |
app = FastAPI(title="Huggingface Safetensor Model Converter to AWQ", version="0.1.0", lifespan=lifespan)
|
17 |
### -------
|
18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
@app.get("/", include_in_schema=False)
|
21 |
def redirect_to_docs():
|
|
|
61 |
raise HTTPException(status_code=500, detail="Failed to convert model")
|
62 |
|
63 |
@app.post("/convert_gpt_q", response_model=None)
|
64 |
+
def convert_gpt_q(request: GPTQConvertionRequest)->Union[FileResponse, dict]:
|
65 |
raise HTTPException(status_code=501, detail="Not implemented yet")
|
66 |
|
67 |
@app.post("/convert_gguf", response_model=None)
|
68 |
+
def convert_gguf(request: GGUFConvertionRequest)->Union[FileResponse, dict]:
|
69 |
raise HTTPException(status_code=501, detail="Not implemented yet")
|
70 |
|
71 |
@app.get("/health")
|
requirements.txt
CHANGED
@@ -5,10 +5,10 @@ torchaudio
|
|
5 |
setuptools
|
6 |
wheel
|
7 |
pydantic
|
8 |
-
fastapi[standard]
|
9 |
-
transformers
|
10 |
huggingface_hub
|
11 |
-
autoawq[kernels]
|
12 |
starlette>=0.46.2
|
13 |
pytest
|
14 |
requests
|
|
|
5 |
setuptools
|
6 |
wheel
|
7 |
pydantic
|
8 |
+
fastapi[standard]>=0.115.12
|
9 |
+
transformers>=4.51.3
|
10 |
huggingface_hub
|
11 |
+
autoawq[kernels]>=0.2.8
|
12 |
starlette>=0.46.2
|
13 |
pytest
|
14 |
requests
|
tests/test_awq.py
CHANGED
@@ -1,7 +1,6 @@
|
|
1 |
import pytest
|
2 |
import requests
|
3 |
from environs import Env
|
4 |
-
from huggingface_hub import login
|
5 |
|
6 |
env = Env()
|
7 |
env.read_env(override=True)
|
@@ -16,6 +15,9 @@ def test_incompatible_model():
|
|
16 |
"hf_push_repo": None,
|
17 |
}
|
18 |
)
|
|
|
|
|
|
|
19 |
assert response.status_code == 400
|
20 |
|
21 |
|
@@ -23,7 +25,7 @@ def test_convert_download():
|
|
23 |
response = requests.post(
|
24 |
f"{env.str('ENDPOINT')}/convert_awq",
|
25 |
json={
|
26 |
-
"hf_model_name": "Qwen/Qwen2.5-
|
27 |
}
|
28 |
)
|
29 |
|
@@ -33,7 +35,7 @@ def test_convert_download():
|
|
33 |
|
34 |
|
35 |
def test_convert_push():
|
36 |
-
model_name = "Qwen/Qwen2.5-
|
37 |
|
38 |
response = requests.post(
|
39 |
f"{env.str('ENDPOINT')}/convert_awq",
|
|
|
1 |
import pytest
|
2 |
import requests
|
3 |
from environs import Env
|
|
|
4 |
|
5 |
env = Env()
|
6 |
env.read_env(override=True)
|
|
|
15 |
"hf_push_repo": None,
|
16 |
}
|
17 |
)
|
18 |
+
|
19 |
+
response.raise_for_status()
|
20 |
+
|
21 |
assert response.status_code == 400
|
22 |
|
23 |
|
|
|
25 |
response = requests.post(
|
26 |
f"{env.str('ENDPOINT')}/convert_awq",
|
27 |
json={
|
28 |
+
"hf_model_name": "Qwen/Qwen2.5-7B-Instruct",
|
29 |
}
|
30 |
)
|
31 |
|
|
|
35 |
|
36 |
|
37 |
def test_convert_push():
|
38 |
+
model_name = "Qwen/Qwen2.5-7B-Instruct"
|
39 |
|
40 |
response = requests.post(
|
41 |
f"{env.str('ENDPOINT')}/convert_awq",
|